I'm having trouble trying to find the best compiling option for amcc440 with fpu. I'm using a genuine fedora core 6 ppc distro (gcc4.1.1). It seems that using mcpu=440 gives better results than -mcpu=440fp, even with floating point code ? What could be wrong on my platform to give such a result ?
Dunno. You need to hunt down more details first: -- Run a profiler (oprofile, gprof) to pinpoint one specific (preferably smallish) piece of code that has a huge slowdown; -- Look at the generated assembler code for that routine to see what's up; -- Try to figure out why that happened, or report back here. Segher
